Document analysis device, document analysis method, and document analysis program

Abstract

The present invention provides a document analysis method capable of presenting MeSH and the like related to a specific physiological activity, work mechanism, functionality, and influence thereof on a living body from a target substance such as a protein, a metabolite, or a nucleic acid obtained from multi-omics analysis by visualizing an analysis result with a unique index.SOLUTION: The method includes at least one of a physiological activity / functionality prediction step of setting the number of selected substances used as an index for predicting a specific physiological activity or functionality and predicting the specific physiological activity or functionality, an enrichment analysis step of performing QPEA registered in a database and quantifying the degree of coincidence between a past research result and a selected substance having a significant difference in an actual research by ES, and a two stage enrichment analysis step of performing two stage enrichment analysis using at least one statistical index selected from ES, NES, and ER.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a document analysis device, a document analysis method, and a document analysis program, and more particularly to a method capable of analyzing health claims identified in a large number of documents. [Background technology]

[0002] Enrichment analysis in metabolomics is an analytical method for statistically associating metabolites with metabolic pathways (see Non-Patent Document 1). Among these, Quantitative Enrichment Analysis (QEA) (see Non-Patent Documents 2 and 3) is known as a quantitative analytical method that utilizes correlation coefficients with objective variables (such as group information), and is widely used as a tool to aid biological interpretation.

[0003] In recent years, various methods have been developed to obtain information from metabolites that will aid in biological interpretation, utilizing the large amount of paper information registered in PubMed and other databases (see Non-Patent Documents 4 and 5, Patent Document 1, etc.).

[0004] However, there are currently no reported methods for using MeSH (Medical Subject Headings) registered in PubMed and the number of papers on multiple selected substances to predict the functionality of selected substances that showed significant differences in actual research, or to quantify the degree of agreement with past research (e.g., reproducibility and novelty).Furthermore, there are no reported methods for extracting MeSH groups with specific functionality or mechanisms of action by combining multiple enrichment analyses. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2022-182568 [Non-patent literature]

[0006]

Outline Building1

Outdoor Tool2

Outdoor Tools3

Outdoor Tools 4

[0007] Conventional methods make it difficult to predict specific physiological activities, mechanisms of action, functionality, and their effects on living organisms. For example, analyzing and predicting health claims for functional foods is technically insufficient, and in particular, visualization of the analysis and prediction results and the prediction accuracy are low.

[0008] Therefore, the first object of the present invention is to provide a document analysis device, a document analysis method, and a document analysis program that can visualize the analysis results using unique indices and present MeSH, etc. related to specific physiological activities, mechanisms of action, functionality, and their effects on living organisms from target substances such as proteins, metabolites, and nucleic acids obtained from multi-omics analysis. The second object of the present invention is to provide a document analysis device, a document analysis method, and a document analysis program that, while conventional QEA uses correlation coefficients with target variables (such as group information), performs QPEA (Quantitative Publication Enrichment Analysis) registered in databases such as PubMed, and can quantify the degree of agreement between past research results and selected substances extracted and selected using specific indicators in actual research using an Enrichment Score (ES). [Means for solving the problem]

[0009] The first document analysis method includes a physiological activity / functionality prediction step in which the number of selected substances (the number of selected substances linked to specific MeSH or set keywords among significant target substances) used as an index for predicting specific physiological activities or functionality is set, and the specific physiological activities or functionality are predicted.

[0010] The second document analysis method involves performing QPEA using data registered in a database (e.g., PubMed) and an enrichment analysis step in which ES quantifies the degree of agreement between past research results and selected substances extracted and selected using specific indicators from actual research. While conventional QEA uses the correlation coefficient with the objective variable (such as group information), this method uses ES, which can measure the degree of agreement with results obtained in previous papers based on the number of papers, allowing us to confirm both reproducibility and novelty at the same time.

[0011] The third document analysis method may include one or more of the first and second document analysis methods, It includes a two-stage enrichment analysis step that uses one or more statistical indicators selected from ES, Normalized Enrichment Score (NES), which is a generalization of ES, and Enrichment Ratio (ER), which is calculated from the number of selected substances and the number of detected substances. This method introduces the concept of MeSH groups, which classify MeSH according to specific physiological activities or functionality, and performs a two-stage enrichment analysis that combines statistical indices (ES, NES, ER) and enrichment analysis of MeSH groups. This significantly improves the accuracy of predicting specific physiological activities and functionality. In other words, by combining multiple statistical indices and multiple enrichment analyses, it becomes a statistical method that can be used not only for predicting health claims in the food industry, but also for interpreting target substances in a wide range of fields. Examples of "prediction of specific physiological activity or functionality" include prediction of health claims, prediction of diseases, and prediction of mechanisms of action.

[0012] The document analysis device of the present disclosure includes: A physiological activity and functionality prediction section sets the number of selected substances (the number of selected substances linked to specific MeSH or set keywords among significant target substances) used as an index for predicting specific physiological activities and functionality, and predicts specific physiological activities and functionality; The enrichment analysis section performs QPEA using the data registered in the database, and quantifies the degree of agreement between the results of past research and the selected substances that showed significant differences in the actual research using ES. a two-stage enrichment analysis unit that performs two-stage enrichment analysis using one or more statistical indicators selected from ES, NES, which is a generalization of ES, and ER, which is obtained from the number of selected substances and the number of detected substances; It has one or more of the following.

[0013] The information processing device At least one processor; a memory for storing instructions executable by said processor; The processor is an information processing device that performs each step of the document analysis method by executing executable instructions.

[0014] Document analysis programs A program for implementing each step of the document analysis method by at least one processor.

[0015] Each element of the device may be configured as an information processing device (e.g., a computer, a server) having a memory, a processor, a software program, a dedicated circuit, firmware, etc. The information processing device may be either on-premise or cloud-based, or a combination of both.

[0016] (effect) (1) The ES calculated from QPEA made it possible to quantify the degree of agreement (reproducibility and novelty) between selected substances extracted and selected using a specific index and previous research. (2) In addition to ES, it is possible to extract more important MeSH from multiple MeSHs by using indicators such as NES and ER. (3) Using the above indicators as a basis, two-stage enrichment analysis enabled the extraction of particularly important MeSH groups. (4) By using the new statistical index of the present invention, it became possible to evaluate the relationship between MeSH and selected substances with high accuracy and extract important information. (5) The analytical method according to the present invention is not limited to metabolomics data, but can also be applied to other analytical data. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 1 illustrates an example of a document analysis apparatus according to a first embodiment. [Figure 2] FIG. 1 is a diagram showing the analysis results of Example 1. [Figure 3] FIG. 10 is a diagram showing the analysis results of Example 2. [Figure 4] FIG. 10 is a diagram showing the analysis results of Example 3. [Figure 5] FIG. 10 is a diagram showing the analysis results of Example 6. [Figure 6] FIG. 10 is a diagram showing the analysis results of Example 7. [Figure 7] FIG. 10 is a diagram showing the analysis results of Example 8. DETAILED DESCRIPTION OF THE INVENTION

[0018] (Embodiment 1: Health Claim Prediction) Embodiment 1 shows an example of predicting health claims among specific physiological activities and functionalities, but the present invention is not limited to predicting health claims.

[0019] The document analysis device 1 includes a physiological activity / functionality prediction unit 11 , an enrichment analysis unit 12 , a two-stage enrichment analysis unit 13 , a storage device 14 , a display device 15 , a communication means 16 , and an input means 17 .

[0020] The physiological activity / functionality prediction unit 11 sets the number of selected substances used as an index for predicting a specific physiological activity or functionality, and predicts the specific physiological activity or functionality. The selected substances are the number of selected substances linked to a specific MeSH or set keyword among significant target substances.

[0021] The enrichment analysis unit 12 performs QPEA registered in the database, and quantifies the degree of agreement between the results of past research and selected substances that showed significant differences in the actual research using ES.

[0022] The two-stage enrichment analysis unit 13 performs two-stage enrichment analysis using one or more statistical indicators selected from ES, NES, which is a generalization of ES, and ER, which is determined from the number of selected substances and the number of detected substances.

[0023] The storage device 14 stores various execution programs, input data, setting data, various databases, output data, etc. for the physiological activity / functionality prediction unit 11, the enrichment analysis unit 12, and the two-stage enrichment analysis unit 13. The storage device 14 may be, for example, various types of memory, or may be on-premise or cloud memory.

[0024] The display device 15 displays various data such as an input screen, an operation screen, and an output screen to be input to the physiological activity / functionality prediction unit 11, the enrichment analysis unit 12, and the two-stage enrichment analysis unit 13. The display device 15 may be, for example, various monitors.

[0025] The communication means 16 communicates various data with external devices. The communication means 16 may be, for example, a wireless communication means or a wired communication means.

[0026] Various data is input through the input means 17. The input means 17 may be, for example, a keyboard, a mouse, a touch panel, or a voice input means.

[0027] The document analysis device 1 or the document analysis program executes the steps of the following document analysis method. (S0-1) From the paper information registered in the PubMed database, the number of papers linking the selected substance with the health claim MeSH is calculated, and a database (D1) is created. In the database (D1), if at least one paper has been reported for each selected substance and each MeSH or set keyword, there is a relationship, and the related information is registered. The keywords are set in advance. The first database creation unit 21 calculates the number of papers linking the selected substance with the health claim MeSH from the paper information registered in the PubMed database, and creates the database (D1).

[0028] (S0-2) A plurality of metabolites related to each MeSH are acquired and organized as a metabolite MeSH classification list. The metabolite MeSH classification list creation unit 22 creates a metabolite MeSH classification list from a plurality of metabolites related to the MeSH.

[0029] (S0-3) Calculate the number of overlaps between multiple pieces of selected substance information that showed significant differences in the actual research and the selected substances in the metabolite MeSH classification list (selection number). The first selection number calculation unit 23 calculates the number of overlaps between multiple pieces of selected substance information that showed significant differences in the actual research and the selected substances in the metabolite MeSH classification list.

[0030] (S0-4) Calculate the Paper Impact (PI), a unique index calculated from the selected substances that showed significant differences in actual research and the number of Health Claim MeSH papers. The PI calculation unit 24 calculates the PI, a unique index calculated from the selected substances that showed significant differences in actual research and the number of Health Claim MeSH papers. The PI is calculated using the following calculation method.

[0031] Let L be the total number of papers in which a specific MeSH and a selected substance that showed a significant difference in actual research are commonly assigned. Then, L = [L1, L2, ..., L k ] the i-th element L i The PI is calculated using the following formula: PI i =(log 10 (Li)-min(log 10 (L))) / (max(log 10 (L))-min(log 10 (L))) However, if there are both a functional ingredient administration group and a placebo group, L can be the sum of L calculated for the functional ingredient administration group and L calculated for the placebo group. The PI value varies depending on the number of papers on the selected substance (a selected substance that showed a significant difference in actual research). Therefore, even if the same MeSH is used, if a substance that has been well studied is selected, the PI value will be higher.

[0032] (S0-5) By visualizing the number of choices and PI using various graphs such as scatter plots, bar graphs, and heat maps, we can predict, for example, the optimal health claims for functional food products. The visualization graph creation unit 25 creates various graphs, such as scatter plots, bar graphs, and heat maps, of the selection numbers and PIs. The created graphs are displayed on the display device 15. The physiological activity / functionality prediction unit 11 analyzes these graphs and predicts specific physiological activities and functionalities from data analysis.

[0033] (Embodiment 2: Two-stage enrichment analysis) The document analysis device 1 or the document analysis program executes the steps of the following document analysis method. (S1) From the paper information registered in the PubMed database, the number of papers in which all selected substances detected in actual research are linked to a specific MeSH is calculated, and a database (D2) is created. The second database creation unit 31 calculates from the paper information registered in the PubMed database the number of papers in which all selected substances detected in actual research are linked to a specific MeSH in common, and a database (D2) is created.

[0034] (S2) A list of the number of papers in the database (D2) sorted in descending order of the number of papers (r={r1, r2, …r N Here, "N" is the number of metabolites included in the database (D2) that have one or more papers for a specific MeSH. The paper number list creation unit 32 creates a paper number list (r={r1, r2, ... r N}).

[0035] (S3) The list (r) is made up of the selected substances with significant differences labeled with 1 and the selected substances without significant differences labeled with 0. N The labeling list creation unit 33 creates a list (Lo={Lo1, Lo2...Lo) in which selected substances that have a significant difference in the list (r) are labeled with 1 and selected substances that do not have a significant difference are labeled with 0. N}). (S4) Let p be the weighting index for the number of papers, and let r be the ith element of list (r) or list (Lo). i , Lo i Then, the total number of papers on selected substances with significant differences is N R is calculated by the following formula (1): The second selection number calculation unit 34 calculates the total number of papers on selected substances that showed significant differences, N R Calculate.

number

[0036] (S5) The number of selected substances for which there is at least one paper for a specific MeSH and which showed a significant difference is N H The second selection number calculation unit 34 calculates the number of selected substances for which there is one or more papers for a specific MeSH and which have a significant difference as N H It is calculated as follows.

[0037] (S6) The proportion of selected substances with significant differences weighted by the number of papers up to the i-th element of list (r) or list (Lo) P hit and the proportion of selected substances with no significant difference, P miss is calculated by the following formulas (2.1) and (2.2). The proportion calculation unit 35 calculates the proportion P of selected substances that showed significant differences. hit and the proportion of selected substances with no significant difference, P miss Calculate.

number

[0038] (S7)ES is [P hit -P miss ] and takes a value between 0 and 1. The closer it is to 1, the more consistent it is with past research and the higher the reproducibility, and the closer it is to 0, the fewer selected substances that match past research and the more novel selected substances there are. The ES calculation unit 36 ​​calculates the proportion P of selected substances that were significantly different. hit and the proportion of selected substances with no significant difference, P miss Calculate ES from

[0039] (S8) Also, list (Lr) is randomly sampled from list (Lo) = {Lr1, Lr2…Lr N The ES calculated using the formula} is defined as ESr. The ES calculation unit 36 ​​also calculates ESr. The ratio (ES / ES0) of the average ESr value ES0 to ES when random sampling is performed, for example, 1000 times, is defined as NES. The NES calculation unit 37 calculates NES. The proportion of ESr values ​​that are equal to or greater than ES among the ESr values ​​calculated by random sampling 1000 times is defined as the p-value of ES. If ESp<0.05, it is set as a statistically high degree of agreement with past research. A different value may be used instead of 0.05. ES and NES are indicators that can calculate the degree of agreement with past research for a specific MeSH, and by calculating the ES and NES for multiple MeSHs, they can relatively evaluate MeSHs that are more closely related to selected substances that showed significant differences.

[0040] (S9) To compare multiple MeSHs, use ER. The average number of selected substances in multiple MeSHs is m s , the average number of detected substances is m d , the number of selected substances in a particular MeSH is n s , the number of detected substances is n d Then, ER is calculated by the following equation (3): The ER calculation unit 38 calculates ER.

number

[0041] (S10) Multiple MeSHs with similar meanings are grouped together, and the multiple MeSHs are compared based on statistical indices (ES, ES p-value, NES, ER) to identify which group contains MeSHs that are biased toward the top, and calculate the ES of the group. The group ES calculation unit 39 calculates the ES of this group. For example, when comparing F100 in health claim prediction, it is possible to identify F24, which is biased towards the top. Here, F24 indicates 24 representative health claims of functional foods, and F100 is a characteristic list that is uniquely derived and linked to MeSH related to F24. The total number of F100 linked to F24 is k. The list when F100 is sorted in descending order of ES is the list (rF) = {rF1, rF2, ... rF kIn list (rF), the list in which F100 that corresponds to a specific F24 is labeled with 1 and F100 that does not correspond is labeled with 0. k} Let p be the weighting index of ES, and rF be the i-th element of list (rF) or list (w). i , w i Then, the total number of ESs for a particular F24 is N RF is calculated using the following formula (4). RF The calculation unit 41 calculates the total N of ES of a specific F24. RF Calculate.

number

[0042] (S11) The number of F100s that correspond to a specific F24 is N HF Let N HF The calculation unit 42 calculates N HF Calculate. The proportion P of F100 that corresponds to a particular F24 weighted by ES up to the i-th element of list (rF) or list (w) hit and the proportion of F100 that does not apply, P miss is calculated by the following formulas (5.1) and (5.2). The ratio calculation unit 43 calculates the ratio P of F100 corresponding to a specific F24 weighted by ES. hit and the proportion of F100 that does not apply, P miss Calculate.

number

[0043] (S12) Enrichment Score F24 (ES(F24)) is [P hit -P miss ] and takes a value between 0 and 1. Normalized Enrichment Score F24 (NES(F24)) is calculated in the same way as NES above. ES F24 The calculation unit 44 calculates ES(F24). F24The calculation unit 45 calculates the NES (F24). The visualization graph creation unit 25 creates various graphs, such as scatter plots such as bubble charts, bar graphs, and heat maps, regarding the relationship between the number of selected substances and PI max, the relationship between PI max and ES, and the relationship between PI max and NES. When an actual study has two groups, a treatment group and a placebo group, the enrichment analysis unit 12 or the two-stage enrichment analysis unit 13 calculates the difference in ES(F24) (diffESES(F24)) or the difference in NES(F24) (difference in diffESNES(F24)) between them, thereby extracting health claims that are more important for distinguishing the two groups. The enrichment analysis unit 12 or the two-stage enrichment analysis unit 13 can correct the disadvantage of the ES of F24, which has only one MeSH, being high by multiplying the difference in ES (F24) by the number of MeSHs of F100 linked to F24. When ESES(F24) of the treatment group is ESES(F24)t and ESES(F24) of the placebo group is ESES(F24)p, diffESES(F24)= ESES(F24)t - ESES(F24)p When ESNES(F24) of the treatment group is ESNES(F24)t and ESNES(F24) of the placebo group is ESNES(F24)p, diffESNES(F24)= ESNES(F24)t - ESNES(F24)p

[0044] The enrichment analysis unit 12 or the two-stage enrichment analysis unit 13 analyzes whether MeSHs included in a specific group of MeSHs are biasedly hit (selected) based on the ES scores. In the list (rF) described in paragraph

[0040] , when F100 is arranged in descending order of ES, it is expressed as follows using the formula [5]. ESES(F24)= max(P hit -P miss ) ESNES(F24)= ESES(F24) / ESES(F24)r Here, ESES(F24)r is the average value of ESES(F24) calculated using a list randomly sampled from list (rF).

[0045] The enrichment analysis unit 12 or the two-stage enrichment analysis unit 13 analyzes whether MeSHs included in a specific group of MeSHs are biasedly hit (selected) based on the ER score. In the list (rF) described in paragraph

[0040] , when F100 is arranged in descending order of ER, it can be expressed as follows using [Equation 5]. ERES(F24) = max(P hit -P miss ) ERNES(F24)= ERES(F24) / ERES(F24)r Here, ERES(F24)r is the average value of ERES(F24) calculated using a list randomly sampled from list (rF).

[0046] The above method is not limited to the hierarchical structure of F24 and F100, but can also be applied to analyses other than predicting health claims about functional foods by grouping MeSH with similar meanings. Here, F24 indicates 24 representative health claims of functional foods, and F100 is a characteristic list that independently derives and links MeSH related to F24.

[0047] In the present invention, unless otherwise specified, the following terms have the following meanings. [Table 1]

[0048] (Example 1: Predicting health claims related to antioxidant effects and sleep improvement from plasma metabolome data) From the Supplementary Data in Reference 1 below, metabolites that changed significantly before and after the intervention test at p<0.05 were obtained in the group whose sleep improved (responders n=7) among those whose staple diet consisted of rice. Using the obtained multiple metabolites with significant differences and the metabolite MeSH classification list of all metabolites measured in the intervention test, an enrichment analysis was performed using Fisher's exact test to calculate the health claim MeSH with the most selections. The analysis results (see Figure 2) showed that antioxidants had a high number of selections and a high PI, predicting that the health claim of "antioxidant effect" would be optimal. This is consistent with the findings in Reference 1, which pointed out that resistance to oxidative stress may have contributed to improved sleep. On the other hand, the health claim for "Sleep," which represents sleep, was in the middle in both the number of selections and PI, and improvements are needed to obtain more clearly targeted health claims. <Reference 1: Minori Koga,, Atsuhito Toyomaki, Yoshinobu Kiso and Ichiro Kusumi “Impact of a Rice-Centered Diet on the Quality of Sleep in Association with Reduced Oxidative Stress:A Randomized, Open, Parallel-Group Clinical Trial”nutrients. (2020) 12, 2926; doi:10.3390 / nu12102926>

[0049] (Example 2: Immunity improvement results based on fecal metabolome data) From the Supplementary Data of Document 2 below, metabolites that significantly varied with p < 0.05 before and after the intervention test were obtained in each of the placebo group (control n = 40) and the mushroom tablet ingestion group (mashroom n = 40). Using the metabolite MeSH classification lists of multiple metabolites with significant differences obtained and all metabolites measured in the intervention test, enrichment analysis was performed using Fisher's exact test, and health claims MeSH with a large number of selections and health claims in ascending order of the p-value of the enrichment analysis were calculated. Figure 3 shows the results of metabolites that were Up-regulate before and after the intervention test. From the analysis results, Antioxidants (antioxidation) had both a large number of selections and a high PI, and it was predicted that the health claim of "antioxidant effect" was the most appropriate. Judging from the p-value of the enrichment analysis, it was suggested that "probiotics (intestinal regulation effect)" and "Intra-Abdominal Fat (visceral fat)" were health claims worthy of attention. <Document 2: Yuichiro Nishimoto1, Junya Kawai, Koichiro Mori, Tenagy Hartanto1, Kaori Komatsu1, Toru Kudo1 and Shinji Fukuda “Dietary supplement of mashrooms promotes SCFA production and moderately associates with IgA production: A pilot clinical study” frontiers. (2023) 09 January; doi 10.3389 / fnut.2022.<1078060>

[0050] (Example 3: Reproducibility and novelty evaluation using the Enrichment Score (ES) calculated from QEA based on the number of papers) From the Supporting Information in Reference 3 below, we identified metabolites with a p<0.05 in repeated measures ANOVA among the serum metabolites of 47 coffee-consuming subjects. To compare with previous coffee-related papers, we created a list of the detected substances and the number of papers containing coffee (MeSH). The ES, ES p-value, and NES were calculated and displayed as an enrichment plot (see Figure 4). The enrichment plot lists metabolites with a large number of papers from left to right on the horizontal axis, with vertical bars indicating the position of metabolites with significant differences. Figure 4 also displays a heat map of each metabolite and the number of papers containing coffee MeSH. ES represents the maximum score. Metabolites located to the left of the ES are considered consistent with previous studies. Analysis revealed high scores (ES: 0.885, ES p-value: 0.004, NES: 1.489), with a 5% significance level. The metabolites with significant differences in this study were statistically confirmed to be consistent with the results of previous coffee-related studies. Figure 4 also visualizes several metabolites that are consistent with previous research, as well as metabolites that have not yet been thoroughly studied due to the limited number of papers published. <Reference 3: MC Cornelis, I Erlund, GA Michelotti, C Herder, JA Westerhuis, J Tuomilehto Metabolomic response to coffee consumption: application to a three-stage clinical trial. J Intern Med. 2018 Jun;283(6):544-557>

[0051] Example 4 From the supplementary data in Reference 4 below, metabolites with a p<0.05 in a two-group comparison test (MW test) between the casein group (n=6) and the whey group (n=6) were obtained. To compare with past antioxidant-related research, a list of the number of papers for Superoxide Dismutase (MeSH), Antioxidants (MeSH), and Oxidative Stress (MeSH) was created for each detected substance. As a result of the analysis, as shown in Table 2, all antioxidant-related MeSH had high ES and NES scores, with a 5% significance level. It was statistically confirmed that the metabolites that showed significant differences in the test were consistent with the results of past antioxidant-related research. Reference 4: Kimitaka Nakazaki, Nobuhiko Nagano, Daichi Katayama, Shoichi Shimizu, Kengo Matsuda, Wataru Tokunaga, Ryoji Aoki, Kazumasa Fuwa, Ichiro Morioka. Body Fat-Reducing Effects of Whey Protein Diet in Male Mice. Nutrients. 2023 May 10;15(10):2263

[0052] [Table 2]

[0053] (Example 5: Extraction of important MeSH based on relative comparison of MeSH using statistical indicators (ES, NES, ER)) The plasma metabolome data in Reference 5 below was obtained from the whhl data in the R package

[11] loadings. A two-group comparison test (Welch's t-test) between the hyperlipidemia group (n = 3) and the hyperlipidemia + drug group (n = 3) determined the top 30 metabolites with significant differences in p-value. ES, ES p-value, ER, and NES were calculated for the MeSH of the F100 health claim prediction package. MeSH with ES p < 0.05 were sorted in descending order of ES for relative comparison. Table 3 shows that several MeSH related to oxidative stress were significant at 5%. The paper suggests that the antioxidant effects of drugs were evident in the hyperlipidemia + drug group. The analysis results suggest that the differences in oxidative stress responses between the two groups are reflected in the MeSH rankings. <Reference 5: Ooga, T., Sato, H., Nagashima, A., Sasaki, K., Tomita, M., Soga T. and Ohashi, Y. (2011) Metabolomic Anatomy of Animal Model Revealing Homeostatic Imbalances in Dyslipidemia. Y. Mol. BioSyst.,7,1217-23.> <Rパッケージ

[11] loadings: Loadings for Principal Component Analysis and Partial Least Squares>

[0054] [Table 3]

[0055] (Example 6: Extraction of important MeSH based on relative comparison of MeSH using statistical indicators (ES, NES, ER)) Using plasma metabolome data from Reference 6, metabolites with p<0.05 were obtained in the sleep improvement group (n=7) and the non-sleep improvement group (n=5) of the rice-based dietary group before and after the intervention study using Wilcoxon signed-rank test. Using the analysis method shown on the left in Figure 5, plotting PI max on the x-axis and F24 as the number of selected substances on the y-axis, the difference in the health claims related to "sleep" between the two groups was unclear. However, using a two-stage enrichment analysis, plotting the difference between the two groups in NES (F24) based on ER on the x-axis and F24 as PI max on the y-axis enabled extraction of "sleep" in the negative direction. This suggests a difference between the two groups in the number of metabolites linked to the MeSH associated with sleep (see Figure 5). <Reference 6: Ooga, T., Sato, H., Nagashima, A., Sasaki, K., Tomita, M., Soga T. and Ohashi, Y. (2011) Metabolomic Anatomy of Animal Model Revealing Homeostatic Imbalances in Dyslipidemia. Y. Mol. BioSyst.,7,1217-23.>

[0056] Example 7 Fecal metabolome data were obtained from the supplementary data in Reference 7 below. A Wilcoxon signed-rank test was performed between the mushroom tablet group (n = 40) and the placebo group (n = 40) before and after the intervention study. The top 20 metabolites with the highest p-values ​​were identified as metabolites with significant differences. Using the analysis method shown on the left in Figure 6, plotting PI max on the x-axis and F24 (number of selected substances) on the y-axis, the health claim for "intestinal regulation" was positioned in the middle, with no significant differences observed compared to other health claims. However, using a two-stage enrichment analysis, plotting F24 with the difference between the two groups in NES (F24) based on ER on the x-axis and PI max on the y-axis, a positive "intestinal regulation" effect was identified. This suggests that the intake of the mushroom tablet, as described in the paper, altered the intestinal environment (see Figure 6). Reference 7: Yuichiro Nishimoto, Junya Kawai, Koichiro Mori, Tenagy Hartanto, Kaori Komatsu, Toru Kudo, Shinji Fukuda. Dietary supplement of mushrooms promotes SCFA production and moderately associates with IgA production: A pilot clinical study. Front Nutr. 2023 Jan 9:9:1078060.

[0057] Example 8 From the supplementary data in Reference 8 below, metabolites with a p<0.05 were obtained in a two-group comparison test (MW test) between the aMCI group (n=58) and the control group (n=57). In the analysis on the left side of Figure 7, when PI max is plotted on the x-axis and F24 as the number of selected substances on the y-axis, the health claim of "memory" was ranked low, but when F24 was plotted in a two-stage enrichment analysis with PI max on the x-axis and NES (F24) based on ES on the y-axis, "memory" came in first (see Figure 7). <Reference 8: Gang Wang, Yi Zhou, Feng-Jie Huang, Hui-Dong Tang, Xu-Hua Xu, Jia-Jian Liu, Ying Wang, Yu-Lei Deng, Ru-Jing Ren, Wei Xu, Jian-Fang Ma, Yi-Nan Zhang, Ai-Hua Zhao, Sheng-Di Chen, Wei Jia Plasma metabolite profiles of Alzheimer's disease and mild cognitive impairment. J Proteome Res. 2014 May 2;13(5):2649-58.> [Explanation of symbols]

[0058] 1 Document analysis device 11. Bioactivity and Functionality Prediction Division 12 Enrichment Analysis Section 13 Two-stage enrichment analysis

Claims

1. A document analysis method including a physiological activity / functionality prediction step of setting a number of selected substances used as an index for predicting a specific physiological activity or functionality, and predicting the specific physiological activity or functionality, The physiological activity / functionality prediction step includes: A process of calculating the number of papers linking a substance with a functional MeSH, which is a MeSH that exhibits specific physiological activity and functionality, from paper information registered in the database, and creating a MeSH classification list that is a list containing multiple substances related to the functional MeSH; a process of calculating the number of overlaps between a plurality of pieces of selected substance information that showed significant differences in actual research and substances in the MeSH classification list, and setting the number of selected substances as the number of selected substances; A process of determining the paper impact (PI), which is an index showing whether the number of papers containing the selected substance is large based on the number of papers containing the functional MeSH and the selected substance that showed a significant difference in actual research; and analyzing the relationship between the number of selected substances and the paper impact (PI) to predict specific physiological activities or functionality. Document analysis methods.

2. It also includes an enrichment analysis step that quantifies the degree of agreement between the results of previous research and the selected substances that showed significant differences in the actual research. The enrichment analysis step includes: A process of creating a database containing the number of papers linked to the functional MeSH from the paper information registered in the database, and creating a labeling list (Lo) in which the database is sorted in descending order of the number of papers to create a paper number list (r) into selected substances with significant differences and substances without significant differences; The proportion of selected substances with significant differences weighted by the number of papers, P, based on the index of weighting of the number of papers (r) or the labeling list (Lo) ｈｉｔ and the proportion of substances with no significant difference P ｍｉｓｓ The proportion of selected substances with significant differences was calculated as P ｈｉｔ and the proportion of substances with no significant difference P ｍｉｓｓ The difference between [P ｈｉｔ -P ｍｉｓｓ ] (the difference is between 0 and 1) to calculate an enrichment score (ES); The closer the enrichment score (ES) is to 1, the higher the degree of agreement with past research, and the closer it is to 0, the fewer selected substances that match with past research and the more novel selected substances are changing. The document analysis method according to claim 1 .

3. The method includes a process of calculating an enrichment ratio (ER), which is an index showing whether the number of selected substances is greater than an expected value, based on the average number of selected substances of the functional MeSH, the average number of detected substances, the number of selected substances of a specific functional MeSH, and the number of detected substances. The document analysis method according to claim 1 .

4. A document analysis method comprising a two-stage enrichment analysis step of performing a two-stage enrichment analysis using one or more statistical indicators selected from the enrichment score (ES) according to claim 2, the normalized enrichment score (NES), and the enrichment ratio (ER) according to claim 3, The two-stage enrichment analysis step includes: A process of calculating an ESr from an enrichment score (ES) calculated using a random list (Lr) obtained by randomly sampling the labeling list (Lo) according to claim 2, and calculating a p-value of the enrichment score (ES) from the proportion of the ESr that has a value equal to or greater than the enrichment score (ES); The average value ESr of the random sampling performed multiple times ０ and the enrichment score (ES) (ES / ES ０ ) to obtain the normalized enrichment score (NES); A process of classifying a set of any plurality of functional MeSHs into one or more groups as a group of functional MeSHs having similar meanings; a list (rF) of the second groups sorted in descending order of one of the enrichment scores (ES), the normalized enrichment score (NES), and the enrichment ratio (ER); a list (w) of the list (rF) in which functional MeSHs of the second group that correspond to specific functional MeSHs of the first group and functional MeSHs of the second group that do not correspond to the specific functional MeSHs of the first group are labeled; and a proportion P of functional MeSHs of the second group that correspond to specific functional MeSHs of the first group weighted by the enrichment score (ES) based on a weighting index of the enrichment score (ES). ｈｉｔ and the proportion P of the functional MeSH in the second group that does not correspond ｍｉｓｓ The difference in the ratio [P ｈｉｔ -P ｍｉｓｓ ] A process of calculating an enrichment score (ES) of the first group from the maximum value (which takes a value between 0 and 1) of The enrichment score (ESr) of the first group is calculated using a list (Lr) obtained by randomly sampling the labeled list (Lo), and the average value ESr of the ESr (ESr of the first group) is calculated. ０ (First group ES ０ ) to the enrichment score (ES) of the first group (enrichment score (ES) of the first group / ES ０ ) to obtain a normalized enrichment score (NES) of the first group; A process of comparing a plurality of functional MeSHs classified into each group based on one or more indicators of the enrichment score (ES), the p-value of the enrichment score (ES), the normalized enrichment score (NES), and the enrichment ratio (ER), thereby identifying which group the functional MeSHs belonging to are biased toward the top; A document analysis method comprising:

5. a physiological activity / functionality prediction unit that sets the number of selected substances used as an index for predicting a specific physiological activity or functionality and predicts the specific physiological activity or functionality; a first database creation unit that creates a database containing the number of papers linking substances with functional MeSHs that are MeSHs that exhibit specific physiological activities and functionality, from paper information registered in the database; and a classification list creation unit that creates a MeSH classification list that is a list including a plurality of substances related to the functional MeSH; a first selection number calculation unit that calculates the number of overlaps between a plurality of pieces of selected substance information that have shown significant differences in actual research and substances in the MeSH classification list, and sets the number of selected substances; a paper impact (PI) calculation unit that calculates a selected substance that has shown a significant difference in actual research and a paper impact (PI) that is an index showing whether the number of papers that include the selected substance is large based on the number of papers that include the functional MeSH; The physiological activity / functionality prediction unit Analyzing the relationship between the number of selected substances and the paper impact (PI) to predict specific physiological activities or functionality. Document analysis device.

6. The enrichment analysis section quantifies the degree of agreement between past research results and selected substances that showed significant differences in actual research. a paper number list creation unit that creates a paper number list (r) in which the database created by the first database creation unit is sorted in descending order of the number of papers; a labeling list creation unit that creates a labeled list (Lo) that separates the selected substances that have significant differences from the substances that do not have significant differences in the paper number list (r); The proportion of selected substances with significant differences weighted by the number of papers, P, based on the index of weighting of the number of papers (r) or the labeling list (Lo) ｈｉｔ and the proportion of substances with no significant difference P ｍｉｓｓ The proportion of selected substances with significant differences was calculated as P ｈｉｔ and the proportion of substances with no significant difference P ｍｉｓｓ The difference between [P ｈｉｔ -P ｍｉｓｓ an ES calculation unit that calculates an enrichment score (ES) from the maximum value of [(the difference) is between 0 and 1]], The enrichment analysis unit The closer the enrichment score (ES) is to 1, the higher the degree of agreement with past research, and the closer it is to 0, the fewer selected substances that match with past research and the more novel selected substances are judged to have changed. The document analysis device according to claim 5.

7. The document analysis device of claim 5 is provided with an ER calculation unit that calculates an enrichment ratio (ER), which is an indicator of whether the number of selected substances is greater than an expected value, based on the average number of selected substances ms, the average number of detected substances md, the number of selected substances ns, and the number of detected substances nd of a specific functional MeSH, which is a MeSH that exhibits multiple physiological activities and functionality.

8. A document analysis device comprising a two-stage enrichment analysis unit that performs two-stage enrichment analysis using one or more statistical indicators selected from the enrichment score (ES) according to claim 6, the normalized enrichment score (NES), and the enrichment ratio (ER) according to claim 7, The two-stage enrichment analysis unit A process of calculating an ESr from an enrichment score (ES) calculated using a random list (Lr) obtained by randomly sampling the labeling list (Lo) according to claim 2, and calculating a p-value of the enrichment score (ES) from the proportion of the ESr that has a value equal to or greater than the enrichment score (ES); The average value ESr of the random sampling performed multiple times ０ and the enrichment score (ES) (ES / ES ０ ) to obtain the normalized enrichment score (NES); A process of classifying a set of any plurality of functional MeSHs into one or more groups as a group of functional MeSHs having similar meanings; a list (rF) of the second groups sorted in descending order of one of the enrichment scores (ES), the normalized enrichment score (NES), and the enrichment ratio (ER); a list (w) of the list (rF) in which functional MeSHs of the second group that correspond to specific functional MeSHs of the first group and functional MeSHs of the second group that do not correspond to the specific functional MeSHs of the first group are labeled; and a proportion P of functional MeSHs of the second group that correspond to specific functional MeSHs of the first group weighted by the enrichment score (ES) based on a weighting index of the enrichment score (ES). ｈｉｔ and the proportion P of the functional MeSH in the second group that does not correspond ｍｉｓｓ The difference in the ratio [P ｈｉｔ -P ｍｉｓｓ ] A process of calculating an enrichment score (ES) of the first group from the maximum value (which takes a value between 0 and 1) of The enrichment score (ESr) of the first group is calculated using a list (Lr) obtained by randomly sampling the labeled list (Lo), and the average value ESr of the ESr (ESr of the first group) is calculated. ０ (First group ES ０ ) to the enrichment score (ES) of the first group (enrichment score (ES) of the first group / ES ０ ) to obtain a normalized enrichment score (NES) of the first group; A document analysis device configured to include a process for identifying which group the functional MeSHs belonging to are biased toward the top by comparing multiple functional MeSHs classified into each group based on one or more indicators of the enrichment score (ES), the p-value of the enrichment score (ES), the normalized enrichment score (NES), and the enrichment ratio (ER).

9. at least one processor; a memory for storing instructions executable by said processor; The information processing device, wherein the processor implements each process of the document analysis method according to claim 1 by executing executable instructions.

10. A document analysis program, A program for implementing each process of the document analysis method according to any one of claims 1 to 4 by at least one processor.

Images